ML-powered email priority classification that gets your support team to the fires that matter — fast.
Customer support teams drown in email. Not all of it is urgent, but finding the emails that are urgent — before an angry enterprise client escalates — is a full-time job in itself.
TriageAI automates that job.
It reads incoming support emails and classifies them into four priority tiers: Urgent, High, Medium, and Low — so your team knows exactly where to look first.
| Metric | Score |
|---|---|
| Overall Accuracy | 82% |
| Urgent Recall | 91% |
| Model | Logistic Regression (class-balanced) |
Urgent recall is the metric that matters most here. Missing an urgent email has real business consequences — a 91% catch rate means fewer fires slip through.
- Data Exploration — Analyse the raw email dataset to understand class distribution, spot the class imbalance (~10% urgent), and identify which raw attributes (subject length, body length, word count) are weak signals on their own.
- Feature Engineering — Transform raw text into meaningful signals: exclamation count, urgent keyword frequency, problem keyword count, and sentiment polarity — the features that actually separate priorities.
- Text length is a weak signal — urgent emails aren't necessarily longer or shorter.
- Exclamation marks matter — urgent emails contain significantly more.
- Keyword frequency is decisive — urgent keyword count is the strongest single predictor.
- Sentiment skews negative — urgent and high-priority emails tend toward more negative polarity.
git clone https://github.com/Nyxox-debug/triageai.git
cd triageai
python -m venv venv
source venv/bin/activate
pip install -r requirements.txttriageai/
├── assets
│ ├── key_features_boxplots.png
│ ├── priority_distribution.png
│ ├── text_length_boxplots.png
│ └── urgency_indicators.png
├── data
│ ├── processed
│ │ ├── emails_with_features.csv
│ │ └── features_complete.csv
│ └── raw
│ └── emails.csv
├── models
│ ├── feature_engineering_metadata.json
│ ├── feature_names.json
│ ├── tfidf_vectorizer.pkl
│ └── train_test_data.pkl
├── notebooks
│ ├── 1-data-exploration.ipynb
│ └── 2-feature-engineering.ipynb
├── README.md
└── requirements.txt
- Python 3.9+
- scikit-learn — modelling & evaluation
- pandas / numpy — data wrangling
- matplotlib — visualisation
- seaborn — visualisation
- TextBlob / NLTK — sentiment & keyword extraction